03 · What You Need to Know
What Belongs in a Useful Preregistration
Start With the Claims You Expect the Study to Support
The easiest mistake is to treat preregistration as a form with boxes to fill rather than as a record of research decisions.
Begin instead with the claims you expect to make. If you intend to claim that an intervention improves a particular outcome, what decisions could materially affect that conclusion? If you intend to test a directional hypothesis, what observation would count as evidence for or against it? If several outcomes will be collected, which one is primary?
This perspective helps determine how much detail is useful. The purpose is not to document every mundane action in advance. It is to specify enough of the inferential path that readers can later distinguish what was planned from what emerged after the results became available.
Research Questions and Hypotheses
If your study tests hypotheses, state them precisely enough that the predicted relationship can be identified. Name the relevant variables, populations or conditions where appropriate, and indicate direction when the hypothesis is genuinely directional.
“Technology affects learning” is too vague to establish much. A more informative hypothesis might specify that students receiving a particular intervention are expected to achieve higher scores on a defined primary outcome than students in a comparison condition.
Not every study requires a conventional hypothesis. Exploratory, descriptive, qualitative, and other designs may instead preregister research questions, objectives, initial expectations, or aspects of the research process. Researchers should not manufacture hypotheses simply to make a study look confirmatory.
Study Design and Conditions
Describe the basic design sufficiently to establish how the research question will be investigated. Depending on the study, this might include experimental conditions, comparison groups, repeated measurements, randomization procedures, blinding, survey waves, observational structure, or other central design features.
This section should answer a practical question: what study did you intend to conduct?
You do not necessarily need to reproduce every procedural detail already documented elsewhere. A detailed research protocol and a preregistration can overlap without being identical. The preregistration should preserve the information needed to understand the decisions whose timing matters.
Sampling Plan and Stopping Rule
Specify how participants, observations, cases, documents, or other units will enter the study. Where relevant, identify the target sample size and explain how it was determined.
The stopping rule can be particularly important. If researchers are free to repeatedly examine results and stop data collection when a desirable pattern appears, the sampling process may become dependent on the observed outcome.
A preregistration might therefore state that recruitment will stop after a specified number of usable observations, on a particular date, after a predefined resource limit, or according to a formal sequential procedure.
Not every study permits a perfectly fixed sample size. The key is to describe the rule you actually intend to use rather than inventing precision that the research context cannot support.
Eligibility and Exclusion Criteria
State which observations will be included and which may be excluded.
For participant research, this may involve eligibility criteria, incomplete responses, failed attention checks, duplicate submissions, implausibly fast completion, protocol violations, or other prespecified conditions. Laboratory studies may have equipment-failure criteria. Secondary-data studies may need rules for missing or unusable records.
Exclusions matter because different defensible rules can produce different analytical samples. Specifying consequential criteria before seeing their effects can make later decisions easier to interpret.
Watch Out
A statement such as “outliers will be removed when necessary” leaves the consequential decision unresolved. If outlier exclusion matters to the analysis, specify how an outlier will be identified and what will happen when one is detected, or state a planned sensitivity analysis when a single rule would be inappropriate.
Variables, Measures, and Outcomes
Identify the variables needed for the primary research questions and explain how they will be operationalized.
If a construct can be measured in several ways, specify which measure will represent it. If a scale contains multiple items, indicate how they will be combined when that decision could affect the analysis. If several outcomes are collected, identify their roles where appropriate, such as primary, secondary, manipulation-check, descriptive, or exploratory outcomes.
These distinctions can prevent a study with many measured outcomes from being reported later as though the most favorable outcome had always been the central one.
Data Processing and Variable Construction
The route from raw data to an analyzed variable may involve substantial discretion.
Researchers may reverse-code items, calculate composite scores, transform variables, recode categories, aggregate repeated observations, handle impossible values, or derive indices from several measures. If such choices are consequential and can reasonably be anticipated, they belong in the preregistration.
Likewise, specify how missing data will be handled when this can be determined prospectively. Different approaches to missingness can affect both the analytical sample and the resulting estimates.
The Analysis Plan Needs More Than the Name of a Statistical Test
Writing “we will use regression” or “data will be analyzed using ANOVA” often leaves important choices unspecified.
A more informative plan identifies which variables enter which analysis, the model specification, relevant interactions, covariates, contrasts, transformations, and the criteria used to evaluate the hypothesis. If several models will be fitted, identify which analysis addresses the primary claim and what the others are intended to accomplish.
For null-hypothesis significance testing, specify the relevant significance threshold where appropriate. If multiple testing adjustments will be used, describe them. If the analysis uses Bayesian inference, equivalence testing, model comparison, machine learning, or another inferential framework, specify the corresponding decision or evaluation criteria that matter for interpretation.
The Open Science Framework's current guidance similarly recommends describing statistical tests, decision criteria, exclusion rules, variable combinations, model form, covariates, and outcomes in a rigorous preregistration.
Link Hypotheses to Analyses
A list of hypotheses followed by a separate list of statistical procedures can still leave ambiguity about which test addresses which prediction.
Where possible, map them explicitly.
Hypothesis Students receiving the intervention will have higher post-test scores than students in the comparison condition after accounting for baseline performance.
Outcome Post-test achievement score calculated according to the prespecified scoring procedure.
Analysis Regress post-test score on study condition and baseline score using the prespecified model.
Evaluation Interpret the estimated intervention effect according to the prespecified inferential criteria and report the estimate with its uncertainty.
This linkage makes the preregistration easier to interpret because the reader can reconstruct the intended path from prediction to evidence.
Specify Foreseeable Contingencies When They Matter
Research plans often depend on conditions that cannot be known in advance. A statistical assumption may fail. Recruitment may be lower than expected. A measure may show inadequate properties. A model may not converge.
You can sometimes anticipate these possibilities with conditional rules:
“If condition A occurs, we will use procedure B; otherwise, we will use procedure C.”
Such rules can preserve flexibility while specifying in advance what will trigger a change. They are often more realistic than pretending that a single analytical route will work under every circumstance.
Not every contingency can be predicted, of course. When something genuinely unexpected occurs, a study can still change after preregistration.
Identify What Is Confirmatory and What Is Exploratory
If your study includes both prespecified hypothesis tests and planned exploratory work, say so.
For example, you might preregister two primary confirmatory analyses while also stating that relationships among several secondary variables will be explored without prespecified hypotheses.
You do not need to preregister the results of an exploration that has not happened yet. Indeed, that would rather defeat the linguistic meaning of “exploration.” What you can document is that exploratory analyses are expected and how you intend to distinguish them from confirmatory claims.
This preserves the complementary roles of exploratory and confirmatory research.
The Right Level of Detail Is Enough to Reduce Meaningful Ambiguity
There is no universal word count that makes a preregistration adequate.
One useful test is to ask whether two competent analysts could read the preregistration and independently reach roughly the same understanding of the primary analysis. If one analyst could reasonably choose one outcome, exclusion rule, model, and covariate set while another could choose entirely different ones, the plan may still leave substantial flexibility.
That does not mean every keystroke must be specified. Excessive detail can create its own problems, particularly when researchers make arbitrary decisions merely because they feel obliged to commit to something before sufficient information exists.
Useful specificity
Clarifies decisions that could materially affect the study's claims and provides workable rules for foreseeable alternatives.
False precision
Commits to arbitrary details that cannot reasonably be determined yet or implies certainty about circumstances the researchers cannot know in advance.
The Appropriate Template Depends on the Research
You do not need to invent a preregistration format from scratch. Registries and research communities provide structured templates.
The Open Science Framework currently offers several registration templates, including a general-purpose OSF Preregistration as well as templates for qualitative research, secondary-data analysis, systematic reviews, simulation studies, eye-tracking research, and other designs.
That variety reflects an important principle: the content of preregistration should follow the methodological logic of the study. A template designed for a conventional confirmatory experiment may be poorly suited to qualitative preregistration or another research design in which decisions emerge differently.
Preregister What You Know, Not What You Wish You Knew
Sometimes a consequential decision genuinely cannot be made yet. Perhaps a later analytical choice depends on information that will become available only after data collection begins.
Do not hide that uncertainty behind vague language. State what remains unresolved, explain why where useful, and specify the decision rule if one can be determined in advance.
A transparent statement of uncertainty can be more informative than an apparently complete plan built on arbitrary commitments.
